Our ancestors were dedicated animal trackers, sometimes following a wounded animal for days at a time. But the development of the Global Positioning System (GPS) has made it much easier to track animals: simply tag an animal and observe its movements from afar.
While this method is better in many ways than following tracks, blood trails, and droppings, using GPS trackers comes with its own complications. In particular, reporting one's location based on GPS more frequently requires higher energy consumption. And if you want to accurately track an animal for a long time, you may need to use a bigger battery than you (or the animal) would prefer.
To address this issue, researchers have developed a new type of GPS tracker that predicts how much an animal may move, accounts for the overall energy budget of the tracking device, and then adjusts energy consumption accordingly. The design, created to track flying foxes in Australia, is described in a study published 5 February in IEEE Transactions on Mobile Computing.
Scientists at CSIRO Ecosystem Sciences in Australia were interested in monitoring the movements of flying foxes (also known as fruit bats), not only because they are important seed dispersers, but also because they serve as a vector for infectious diseases, including Hendra virus in Australia and Nipah virus in Asia.
“Key to understanding and managing these animals is to understand how they utilize landscapes and how they interact with other disease hosts, which requires a fine-grained understanding of their movement,” explains Philipp Sommer, who helped build the new system while serving as a postdoctoral fellow at the CSIRO Distributed Sensing Systems Group.
But more accurate tracking requires taking more frequent GPS samples, which ultimately means the device consumes more energy. Therefore, Sommer and his colleagues wanted a GPS tracking device that could be conservative with its sampling when an animal is not moving much, to optimize the device’s energy budget.
The new system developed by Sommer and his colleagues involves three layers that collectively allow for more accurate localization on a device small enough to be fitted to a flying fox or other such animal. It accounts for the general movement patterns of the species being studied, and even the individual being tracked. Just as some humans are more likely to commute to and from work around 9 AM and 5 PM, flying foxes have similar foraging schedules that are somewhat predictable.
The new GPS system learns to recognize and account for these patterns. First, the system is trained offline using data from other animals of the same species, which gives it a population baseline. “Once deployed in the field, the system can learn the parameters online that are specific to the individual animal, based on the activity and energy harvesting of recent days,” Sommer explains.
Did something unusual happen to disrupt the normal behavior of our flying fox friend? No problem. The system detects the anomaly and falls back to a population-based estimate for the remainder of the flying fox’s expedition.
The second layer of the system is an energy-awareness feature responsible for keeping track of the energy consumed and harvested by the system (energy monitoring) and to estimate the energy resources that will be available in the near future (energy prediction). In a third layer, a software program controls sensors that are linked to the hardware platform. Lastly, at the core of their approach, a scheduling algorithm decides the optimal time to obtain the next GPS sample. And if the animal isn’t moving, there’s no need to take a sample at all.
Sommer says, “Our evaluation results have shown that our approach can significantly increase positioning accuracy for a dynamic energy budget. This means that we are able to capture the trajectory of the animal with less errors or gaps compared to existing approaches.”
When the team compared their method to four other sampling approaches, they found that the systems that employ a fixed GPS sampling interval once the animals begin to move had median tracking errors as high as 146.9 meters. In contrast, the adaptive information-based GPS sampling approach developed by Sommer and his colleagues reduced the median tracking error to 21.6 meters, which comes close to the performance of the optimal offline algorithm with a median tracking error of 15.8 meters.